A Labor/Leisure Tradeoff in Cognitive Control

Wouter Kool and Matthew BotvinickPrinceton University

Daily life frequently offers a choice between activities that are profitable but mentally demanding(cognitive labor) and activities that are undemanding but also unproductive (cognitive leisure). Althoughsuch decisions are often implicit, they help determine academic performance, career trajectories, andeven health outcomes. Previous research has shed light both on the executive control functions thatultimately define cognitive labor and on a “default mode” of brain function that accompanies cognitiveleisure. However, little is known about how labor/leisure decisions are actually made. Here, we identifya central principle guiding such decisions. Results from 3 economic-choice experiments indicate that themotivation underlying cognitive labor/leisure decision making is to strike an optimal balance betweenincome and leisure, as given by a joint utility function. The results reported establish a new connectionbetween microeconomics and research on executive function. They also suggest a new interpretation ofso-called ego-depletion effects and a potential new approach to such phenomena as mind wandering andself-control failure.

Keywords: decision making, mental effort, behavioral economics, cognitive control

Imagine a high-school student sitting at her home computer ona school night. Moment by moment, she faces a recurring decision:Should I focus on my calculus homework, or should I take a breakto daydream or watch online videos? The student’s decision isessentially one between cognitive labor and cognitive leisure,mental work and mental rest. In the language of cognitive psy-chology, the labor in question involves the effortful engagement ofexecutive control: a set of functions, dependent on the prefrontalcortex, that configure information-processing resources for theexecution of computationally intensive, nonroutine tasks (Miller &Cohen, 2001). The student’s moment-by-moment decision is thusbetween a gainful activity that demands robust cognitive controland alternatives that do not.

Cognitive labor/leisure decisions play an obvious role in deter-mining academic and career performance. They may also havesafety implications in settings such as air-traffic control or power-plant operation; neuroimaging research has shown that lapses ofattention and response errors are preceded by a shift in activationfrom dorsal cortical areas underlying cognitive control to a default-mode network that is characteristically most active at rest (Eicheleet al., 2008; Weissman, Roberts, Visscher, & Woldorff, 2006).

Furthermore, cognitive labor/leisure decisions may go awry inclinical disorders like depression or addiction, leading to maladap-tive decision making (Baler & Volkow, 2006; Murphy et al., 2001;van der Plas, Crone, van den Wildenberg, Tranel, & Bechara,2009; Wenzlaff & Bates, 1998). Given these and other consider-ations, it is important to understand how such decisions are made.

Recent behavioral and neuroscientific research has indicatedthat the exertion of cognitive control is intrinsically costly oraversive and that control is only robustly recruited in the presenceof relevant incentives (Jimura, Locke, & Braver, 2010; Kool,McGuire, Rosen, & Botvinick, 2010; McGuire & Botvinick,2010). Cognitive labor/leisure decisions thus appear to involve acost–benefit analysis, which weighs the payoff from cognitivework against its inherent cost.

Because cost–benefit analyses typically result in binary go/no-go decisions, it is tempting to think of cognitive labor/leisuredecisions as involving a categorical, all-or-none choice betweenlabor and leisure. The aim of the present work was to evaluate amore complex but also more interesting account, according towhich decision making is based on preferences for a particularbalance or mixture of cognitive labor and leisure.

A formal framework for understanding labor/leisure tradeoffs isprovided by economic models of labor supply (Nicholson & Sny-der, 2008). Such models address the question of how workerschoose their preferred hours: Given a particular hourly wage and a“budget” of available hours per week, how does a worker chooseto allocate time between work and leisure? Classical labor supplytheory proposes that workers value both income and leisure andthat overall value or utility is a joint function of both. A graphicaldepiction of a representative utility function is presented in Figure1A. Here, the x-axis indicates leisure time, the y-axis total incomefrom labor, and the z-axis utility. The utility function takes theform of a surface, which attaches a subjective value to each pairingof income with leisure time (the lower portion of the figure showsthis as a contour plot). A central proposition of labor supply theory

This article was published Online First December 10, 2012.Wouter Kool and Matthew Botvinick, Department of Psychology and

Princeton Neuroscience Institute, Princeton University.This research was supported by National Institute of Mental Health

Grant MH062196 and a grant from the John Templeton Foundation. Theopinions expressed in this publication are those of the authors and do notnecessarily reflect the views of the John Templeton Foundation. We thankN. I. Córdova, J. Kaplan, and J. Prabhakar for research assistance. Theauthors declare that they have no conflict of interest with respect to theirauthorship or the publication of this article.

Correspondence concerning this article should be addressed to WouterKool, Department of Psychology, Green Hall, Princeton University,Princeton, NJ 08540. E-mail: wkool@princeton.edu

Thi

sdo

cum

ent

isco

pyri

ghte

dby

the

Am

eric

anPs

ycho

logi

cal

Ass

ocia

tion

oron

eof

itsal

lied

publ

ishe

rs.

Thi

sar

ticle

isin

tend

edso

lely

for

the

pers

onal

use

ofth

ein

divi

dual

user

and

isno

tto

bedi

ssem

inat

edbr

oadl

y.

Journal of Experimental Psychology: General © 2012 American Psychological Association2014, Vol. 143, No. 1, 131–141 0096-3445/14/$12.00 DOI: 10.1037/a0031048

131

is that this utility function is quasiconcave, favoring combinationsof labor and leisure that are relatively balanced.

Economic labor supply theory provides a simple and compellingaccount of how labor/leisure decisions are made, given this form ofutility function. Note that the worker’s available hours and wagedefine a range of attainable income/leisure combinations, referredto in labor supply theory as the budget constraint. A sample budgetconstraint is shown in Figure 1A, appearing as a diagonal line inthe lower part of the figure and as a gray plane in the surface plot.According to labor supply theory, the worker selects, from amongall feasible income/leisure combinations, the one yielding thegreatest utility. In Figure 1A, this optimum is marked in red. As thefigure illustrates, the concave utility function leads to decisionsthat favor a balance between income and leisure, resisting a strongbias toward either one or the other.

Labor supply theory predicts interesting, nonintuitive shifts inchoice behavior with variations in wage. For example, Figure 1B

diagrams the effect of an income-compensated wage decrease.Here, the worker is given an initial baseline wage (in blue) andselects the income/leisure combination associated with maximalutility (corresponding, in the diagram, to the point of contactbetween the budget constraint line and the highest isoutility curvewith which it intersects). Next, a wage reduction is imposed (inred), but the worker is also given a “freebie” payment up front.This unearned income precisely compensates for the wage reduc-tion, in the sense that if the worker chooses the same work hoursas she did before the wage change, her total income (including thefreebie payment) will also remain unchanged (the blue and redbudget constraints cross at the initial allocation). Despite theavailability of this status quo ante income/leisure combination,the geometry of the utility surface in Figure 1A dictates that theworker will reduce her hours, settling for a smaller total income.To see this, note that the new optimal point lies on an indifferencecurve associated with higher utility than the previous equilibrium.

Baseline

A B

C

After decrease

After increase

Inco

me

Leisure

Inco

me

Leisure

Baseline

Util

ity

T

T

0

0

0

0

Figure 1. A: Above: A concave utility surface attaching a scalar value to each pairing of income with leisuretime. Shown in black are isoutility or indifference curves. Below: A contour map, projected from the indifferencecurves above. The diagonal line represents a budget constraint, covering the range of income/leisure combina-tions open to a worker facing T available hours and wage w. The gray rectangle above projects this constraintline back up through the three-dimensional utility surface, with the intersection forming an arc. In both top andbottom plots, the red marker indicates the income/leisure combination with the highest utility in the feasible set.B: An income-compensated wage decrease. The initial, baseline wage, together with the total available time T,defines the initial budget constraint line (blue). The worker chooses the point along this line that maximizesutility (blue point, with associated indifference curve shown in gray). The wage reduction results in a morehorizontal budget constraint (diagonal red line). However, it comes along with unearned income (vertical redline), which causes the new budget constraint to intersect the income/leisure combination originally chosen (bluepoint). Although this combination is thus still available after the wage change, the worker also has access to ahigher utility combination (red point, with associated isoutility contour) involving increased leisure and reducedincome. C: An income-compensated wage increase, illustrated in the same fashion. Here an initial unearnedincome (vertical blue line) plus an initial wage together define a budget constraint (diagonal blue line). Whenthe wage is increased, the unearned income is reduced (vertical green line), making the originally chosenincome/leisure combination (blue point) still available. However, the worker now has access to a higher utilitycombination involving less leisure and greater income (green point).

Thi

sdo

cum

ent

isco

pyri

ghte

dby

the

Am

eric

anPs

ycho

logi

cal

Ass

ocia

tion

oron

eof

itsal

lied

publ

ishe

rs.

Thi

sar

ticle

isin

tend

edso

lely

for

the

pers

onal

use

ofth

ein

divi

dual

user

and

isno

tto

bedi

ssem

inat

edbr

oadl

y.

132 KOOL AND BOTVINICK

An opposite but otherwise analogous effect is induced by anincome-compensated wage increase (see Figure 1C). Here, partici-pants are initially given a freebie payment and baseline wage (in blue).In a second session, an increase in wage is paired with a reduction inunearned income (in green). Although the worker once again hasaccess to the same income/leisure combination that she chose beforethe wage change, labor supply theory predicts that the worker willsacrifice leisure time in order to attain a larger total income.

A question that often arises when first encountering a descrip-tion of these effects is, Why bother with the income compensation?Why not consider changes in work allocation under simple in-creases or reductions in wage? It may appear obvious that laborshould increase and leisure decrease in response to an increase inwage. However, as it turns out, this sort of simple monotonicrelationship between wage and work is not at all predicted by laborsupply theory. It is in fact an important corollary of the theory thatincreases in wage can, under certain circumstances, yield a reduc-tion in work. Empirical findings support this aspect of labor supplytheory. Nonmonotonic relations between wage and labor supplyhave been observed in behavioral–economic studies involvingboth human workers (Bigelow & Liebson, 1972) and animalworkers (Collier & Jennings, 1969; Green, Kagel, & Battalio,1987). Income compensation is thus a critical measure, since ityields unambiguous predictions: Labor supply theory stronglypredicts that income-compensated wage increases (decreases)should be followed by increases (decreases) in effort supplied.

When such effects are empirically observed, they constitute directevidence for preferences that balance income against leisure. Income-compensated wage manipulations therefore offer a powerful tool forprobing labor/leisure decision making. In recognition of this, suchwage manipulations have been applied in numerous experimental andfield studies on labor markets (e.g., Charness & Kuhn, 2010; Dick-inson, 1999; Fehr & Goette, 2007), as well as in animal conditioningresearch (Chen, Lakshminarayanan, & Santos, 2006; Kagel, Battalio,& Green, 1995). In the present work, we leveraged the same approachto test for balance-oriented preferences in human cognitive control.Our general hypothesis was that cognitive labor/leisure decisionsoperate like the labor/leisure decisions studied in labor economics,tending toward a balance between income and leisure. Based on this,our specific prediction was that cognitive labor/leisure decisionsshould respond to income-compensated wage manipulations as dic-tated by labor supply theory.

In what follows, we report three experiments. In all three,participants freely allocated time between a highly demandingcognitive task, which paid a wage (labor), and a nondemandingtask, which did not (leisure). The leisure task was closely matchedto the labor task on stimulus and response characteristics, demandsfor physical effort, and intrinsic interest; the primary factor thatdifferentiated the two tasks was cognitive demand, the key ingre-dient distinguishing cognitive labor from cognitive leisure as weintend these terms. In each experiment, we measured changes intime allocation induced by particular wage manipulations, testingpredictions derived from the labor supply theory model (see Fig-ures 1B and 1C).

Experiment 1

Our first experiment imposed an income-compensated wagereduction. As illustrated in Figure 1B, the labor supply theory

account predicts that this manipulation will induce participants toforego a previously selected labor/leisure combination, choosinginstead to sacrifice income by reducing time in cognitive labor.

Method

Participants. Thirty-three participants (23 female) from thePrinceton University community provided consent and receivedcourse credit or a nominal payment. All procedures were approvedby Princeton University Institutional Review Board.

Procedure. Participants were asked to spend 40 min perform-ing a combination of two computer-based tasks: a high-demandtask that placed heavy demands on executive function and alow-demand task that did not. Both tasks consisted of the sequen-tial presentation of photographs of humans’ faces (one face every1,750 ms). In the low-demand task, if this showed a child, theparticipant was to depress a joystick button (20% of all trials). Tominimize demands for cognitive control, a tone would alwaysaccompany the presentation of child images. The high-demandtask involved identical stimuli and timing (including the toneaccompanying children’s faces) but called for a button press whenany face matched the one presented three steps earlier (also 20% ofall trials). Note that this task involves the active maintenance andupdating of humans’ faces in working memory and thereforerequires cognitive control (Braver et al., 1997).

Stimuli for the two tasks were presented on opposite sides of acomputer screen, and participants chose between the tasks bydirecting a joystick to the relevant side. Instructions indicated thatparticipants could allocate their time between the tasks howeverthey pleased and could switch from one to the other whenever andhowever often they wished. Importantly, participants were paid awage, denominated in pieces of M&M’s candy (Mars Incorpo-rated, Hackettstown, NJ), for each trial dedicated to the high-demand task. The low-demand task, in contrast, yielded no suchpay. A number indicating total accrued reward for the presentsession was always visible at the top of the computer monitor, andit increased with each trial presented during the high-demand taskaccording to the size of the wage. Participants were explicitlyinstructed that payment of the wage was not contingent on theaccuracy of their responses on the high-demand task but only onthe number of trials dedicated to this task. To further discourage aspecial focus on error monitoring, no feedback on response accu-racy was provided. Participants were simply instructed, at theoutset of the experiment, to do their best on both tasks. Eachsession of the experiment began with a short practice session,allowing familiarization with both tasks and with the payoff re-gime, and it concluded with delivery of earned income. Becausethe tone accompanying children’s faces was also played during thehigh-demand task, it was possible that participants used this toneto more easily identify target trials with children’s faces. However,paired t tests on the data of this and the following experimentshowed that accuracy was not higher for target trials with chil-dren’s faces than for target trials with adults’ faces (ps � 0.10).

The experiment involved two sessions, separated by 1 to 2weeks. In Session 1, all participants were accorded the samebaseline wage (0.029 pieces of candy for each high-demand trial)and cumulative earnings (income), and the time allocated to thehigh- and low-demand tasks—time in labor and (relative) leisure—was recorded. In Session 2, an income-compensated wage re-

Thi

sdo

cum

ent

isco

pyri

ghte

dby

the

Am

eric

anPs

ycho

logi

cal

Ass

ocia

tion

oron

eof

itsal

lied

publ

ishe

rs.

Thi

sar

ticle

isin

tend

edso

lely

for

the

pers

onal

use

ofth

ein

divi

dual

user

and

isno

tto

bedi

ssem

inat

edbr

oadl

y.

133COGNITIVE LABOR/LEISURE DECISIONS

duction was imposed: Participants were explicitly told that thewage from Session 1 was reduced by 50% but that they alsoreceived a freebie candy payment up front (though they were askednot to consume any of it until after completion of the experiment).This unearned income was calibrated separately for each partici-pant so as to make it possible for the participant to allocate exactlythe same amount of time to labor as in Session 1 and accumulateprecisely the same total income (see Figure 1B). Our prediction,drawn from labor supply theory, was that the wage change wouldinduce participants to forgo this option, settling instead for asmaller total income along with increased leisure time.

Because the wage level in Session 1 was established a priori,without information about each participant’s baseline preferences,we anticipated that the wage might induce some participants toallocate the large majority of this session either to labor or toleisure. It is important to note that such events would not contradictany prediction from labor supply theory. Despite the balanceprinciple it implies, labor supply theory does allow that particularincentive regimes may yield an optimum strongly favoring eitherlabor or leisure. Furthermore, some participants may have consid-ered their time budget (as defined in the introduction) to extendbeyond the 40 min of the testing period, leading to choices thatbalanced within-session labor against pre- or postsession leisure.Notwithstanding these points, we were concerned that cases ofnearly all-or-none time allocation might reduce the sensitivity ofour design by introducing a ceiling or floor effect. For this reason,only participants who allocated more than 5% of each session (i.e.,2 min) to both labor and leisure (n � 20) were included in the finaldata set.1 Only participants satisfying this criterion in Session 1were invited to complete Session 2.

Analyses. The central prediction was that leisure time wouldincrease (and total income therefore decrease) from Session 1 toSession 2. Leisure time in Sessions 1 and 2 was compared by wayof a paired t test, excluding two participants whose differencebetween the two sessions fell greater than two standard deviationsbeyond the sample mean. Additional analyses, aimed at verifyingtask compliance, are described in conjunction with results.2

Results and Discussion

To confirm compliance with task instructions on both high- andlow-demand tasks, we computed response accuracies and d= indi-ces for each. Results, summarized in Table 1, indicated that par-ticipants made an adequate effort to perform both tasks as in-structed, responding well above chance levels (single-sample ttests, ps � .001). In both sessions, d= scores and accuracy werelower for the high-demand task than for the low-demand task(paired t tests, ps � .001), validating our demand manipulation.

Labor/leisure allocations were positively correlated between thetwo sessions (r � .72, p � .001), indicating that preferences wererelatively stable over the intersession interval. However, the wagemanipulation had a significant effect. Mean leisure time and in-come for Sessions 1 and 2 are shown in Figure 2A. On averageacross participants, the income-compensated wage decrease trig-gered a 29% increase in leisure time and a 6% decrease in income,a statistically significant shift in the predicted direction, t(17) �2.48, p � .05, Cohen’s d � 0.58.

Experiment 2

The results of Experiment 1 supported the idea that effort-baseddecision making is governed by a preference for balance betweencontrol and leisure. As predicted by labor supply theory, anincome-compensated wage reduction led participants to reducetime in labor, settling for less income, even though it was possibleprecisely to recreate an earlier chosen labor/leisure combination.Experiment 2 tested the complementary prediction, by imposing anincome-compensated wage increase. As articulated in Figure 1C,labor supply theory here predicts that participants will here forgoa previously selected labor/leisure combination, choosing to in-crease labor in order to accrue a larger income.

Method

Participants. Twenty-one members (15 female) of the Prince-ton University community participated, providing informed con-sent and receiving course credit or a nominal payment.

Design and procedure. The procedure was largely identicalto that in Experiment 1. However, here, in Session 1, participantswere accorded a lower baseline wage and received an up-frontfreebie payment. In Session 2, the wage was doubled and partic-ipants received an up-front payment smaller than the payment inSession 1. This new freebie payment was calibrated so as to makeit possible for each participant to allocate exactly the same amountof time to labor as in Session 1 and to accumulate precisely thesame total income (see Figure 1C). Again, our prediction fromlabor supply theory was that the wage change would induceparticipants to forgo this option, causing them instead to sacrificeleisure time in favor of a larger total income.

As in Experiment 1, measures were taken to mitigate floor andceiling effects. Here, this was accomplished by focusing analysison participants (n � 16) who allocated at least 5% of each sessionto both labor and leisure tasks.3 In contrast to Experiment 1, allparticipants completed both Sessions 1 and 2.

Analyses. The central prediction was that mean leisure timewould decrease (and total income thus increase) from Session 1 toSession 2. This was tested by a paired t test, excluding oneparticipant whose difference between sessions fell greater than twostandard deviations beyond the sample mean. Additional analysesare once again described in conjunction with results.

Results and Discussion

As summarized in Table 1, response accuracies again indicatedcompliance with task instructions, far exceeding chance levels(single-sample t tests, ps � .001). As in Experiment 1, participantsshowed better performance for the low-demand task than for thehigh-demand task (paired t tests, ps � .001).

1 There were 11 participants who allocated more than 95% of the sessionto the high-demand task and two participants who allocated greater than95% to the low-demand task.

2 All t test analyses reported in the present article were repeated using anonparametric procedure (signed-ranks test), yielding similarly significantresults in every case.

3 There were four participants who allocated more than 95% of thesession to the high-demand task and one participant who allocated greaterthan 95% of the time to the low-demand task.

Thi

sdo

cum

ent

isco

pyri

ghte

dby

the

Am

eric

anPs

ycho

logi

cal

Ass

ocia

tion

oron

eof

itsal

lied

publ

ishe

rs.

Thi

sar

ticle

isin

tend

edso

lely

for

the

pers

onal

use

ofth

ein

divi

dual

user

and

isno

tto

bedi

ssem

inat

edbr

oadl

y.

134 KOOL AND BOTVINICK

Again as in Experiment 1, time allocation was positively cor-related across Sessions 1 and 2 (r � .81, p � .001). But once again,the wage manipulation induced significant changes in task choice.Mean leisure time and income for Sessions 1 and 2 are shown inFigure 2B. Across participants, the income-compensated wageincrease triggered a 13% decrease in leisure time and a 9%increase in income, once again a statistically significant shift in thepredicted direction, t(14) � 2.36, p � .05, Cohen’s d � 0.61. This

effect remained significant even when the analysis included par-ticipants who dedicated less than 5% of one session to either laboror leisure, t(19) � 3.29, p � .004, Cohen’s d � 1.51.

Experiments 1 and 2: Additional Analyses

One potential concern in interpreting the time-allocation resultsfrom Experiments 1 and 2 is that participants might have based

Table 1Average Performance Scores (and Standard Errors) for the High-Demand and Low-Demand Tasks in Experiments 1–3

Measure

Experiment 1 Experiment 2 Experiment 3

Session 1 Session 2 Session 1 Session 2 Practice

d=Low demand 4.17 (0.10) 4.06 (0.10) 4.28 (0.15) 3.60 (0.26) 1.69 (0.11)High demand 1.82 (0.08) 1.88 (0.13) 1.63 (0.10) 1.63 (0.11) 3.44 (0.04)

Accuracy (%)Low demand

Target 98 (0.4) 96 (0.9) 95 (2.2) 85 (5.3) 100 (0)Nontarget 99 (0.2) 99 (0.1) 99 (0.1) 99 (0.3) 99.6 (0.2)

High demandTarget 58 (3.2) 54 (4.5) 49 (4.0) 42 (4.7) 61 (4.9)Nontarget 94 (0.8) 96 (0.6) 95 (0.4) 96 (0.4) 92 (1.7)

Reaction time (ms)Low demand 549 (24) 552 (23) 599 (32) 674 (51)High demand 753 (20) 789 (30) 824 (28) 844 (38)

44

54

11 16

Baseline

BaselineAfter increase

27

32

14 22

After

Baseline

Baseline

After increase

Leisure (min)

Inco

me

(pcs

)

decrease

After decrease

A B

C D

25

30

8 17

22

28

8 16Leisure (min)

Inco

me

(pcs

)

Leisure (min)

Inco

me

(pcs

)

Leisure (min)

Inco

me

(pcs

)

b

a

c

d

Figure 2. A: Results of Experiment 1 (compare with Figure 1B). Leisure time is in minutes, and income inpieces of candy. Red and blue lines indicate mean budget constraints. Error bars indicate within-subject SEM(note that horizontal error bars are simply a scaled version of the vertical bars). B: Results of Experiment 2(compare with Figure 1C). C: Results for wage-decrease trial pairs in Experiment 3 (compare with Figure 1B).D: Results for wage-increase trial pairs in Experiment 3 (compare with Figure 1C). Note that the higher baselineincome in this condition reflects relatively high unearned income (see Table 2).

Thi

sdo

cum

ent

isco

pyri

ghte

dby

the

Am

eric

anPs

ycho

logi

cal

Ass

ocia

tion

oron

eof

itsal

lied

publ

ishe

rs.

Thi

sar

ticle

isin

tend

edso

lely

for

the

pers

onal

use

ofth

ein

divi

dual

user

and

isno

tto

bedi

ssem

inat

edbr

oadl

y.

135COGNITIVE LABOR/LEISURE DECISIONS

their choices exclusively on the proffered wage, ignoring unearnedincome. As noted in the introduction, it is not necessarily the casethat simple (uncompensated) increases in wage yield increases inwork. Indeed, in the context of cognitive tasks, past empirical workhas failed to reveal any simple relationship between incentivemagnitude and effort investment, as reflected in task performance(Bonner & Sprinkle, 2002; Camerer & Hogarth, 1999). Neverthe-less, it is worth asking whether the shifts in effort allocationobserved in Experiments 1 and 2 might reflect a reaction to simplewage changes alone, in isolation from unearned income. To eval-uate this, we compared leisure time between Experiment 1, Ses-sion 1, and Experiment 2, Session 2—sessions that involved equalwages but different freebie payments (see Panels A and D inFigure 2). We also compared leisure time in Experiment 1, Session2, with leisure time in Experiment 2, Session 1—since here againwages were equal but unearned income differed (see Panels B andC in Figure 2). In both cases, leisure time differed across therelevant sessions (two sample t tests, ps � .05). This indicates thattime allocation was not based on wage alone. Rather, participantsappear to have based their labor/leisure decisions jointly on wageand unearned income.

We interpret the results of Experiments 1 and 2 in terms of effortcosts, or the intrinsic cost of cognitive control. It is partly relieffrom these costs, we propose, that underlies the inherent utility ofleisure, as it is termed in labor supply theory. However, one mightspeculate that what is aversive in the high-demand task is not itscognitive demand, per se, but rather its association with frequentresponse errors. Of course, our participants were given no explicitfeedback on response accuracy and knew that wage delivery wasnot contingent on performance. More important, data from anotherstudy (Kool et al., 2010) show that human decision makers avoidcognitive demand even in the absence of error commission. Erroravoidance thus seems unlikely to be the only factor underlying thepresent results. However, to evaluate the potential role of errors,we conducted two additional analyses. First, based on the recom-mendation of an anonymous reviewer, we tested whether the wageeffects from Experiments 1 and 2 would remain significant aftercontrolling for shifts in response accuracy on the high-demand taskfrom Session 1 to Session 2. Repeated-measures analyses of co-variance, using intersession change in d= as a covariate, revealedintact effects of wage on leisure time for both experiments. Sec-ond, within each experimental session, we examined whetherparticipants tended to switch away from the high-demand taskafter the occurrence of errors. We considered trials where theparticipant switched from the high- to the low-demand task (switchtrials) and trials where the participant remained on the high-demand task (stay trials), comparing these for mean accuracy overthe preceding run of high-demand trials. Repeated-measures anal-yses of variance revealed no significant difference in either sessionof Experiments 1 and 2 for history lengths up to 11 trials. At longerhistory lengths (up to 30 trials), with no correction for multiplecomparisons, an effect appeared in only one of four sessions(Experiment 2, Session 2, History Lengths 12, 13, and 16–30). Insum, taken together with previous research, the data do not com-pellingly support an account based entirely on error avoidance.Further evidence along these same lines is presented in conjunc-tion with the next experiment.

Experiment 3

Experiments 1 and 2 provide support for an income/leisuretradeoff in cognitive control allocation, analogous to the onedescribed in economic labor supply theory. However, one mightsuspect that the tradeoff evident in those experiments arises as anemergent feature of cognitive control dynamics. An influential lineof research has suggested that cognitive control is subject tofatigue, or resource depletion (Muraven, Tice, & Baumeister,1998). If this is correct, then perhaps participants in Experiments1 and 2 allocated time between tasks by simply responding, mo-ment by moment, on the basis of their current level of fatigue,resting when highly depleted and returning to the high-demandtask when replenished.

If this were the whole story, then the effects observed in our firsttwo experiments, during online effort allocation, should disappearwhen allocation decisions must be made prospectively. The pres-ent experiment examined effort allocation in precisely this setting.

Method

Participants. Nineteen members (12 female) of the PrincetonUniversity community participated, providing informed consentand receiving either course credit or a nominal payment.

Design and procedure. The experiment started with 2 min ofpractice on both the high- and low-demand tasks used in Experi-ments 1 and 2. Next, prior to entering a 20-min task-performanceperiod, participants used a graphical interface to allocate timeprospectively between the high- and low-demand tasks (see Figure3). Each participant made 32 prospective allocation decisions,involving a variety of wages and up-front payments, on the un-derstanding that one of their choices would be chosen randomlyand used to program stimulus presentation during the later task-performance period. Embedded in the choice task was a set ofrandomly interleaved trial pairs. One member of each pair servedas a baseline, while the other implemented an income-compensated wage decrease or increase (see Table 2). On half ofall trials, the awarded candy pieces were M&M’s, and on the otherhalf Reese’s Pieces (Hershey Company, Hershey, PA). Upon com-pletion, participants received the amount of candy selected on arandomly chosen time-allocation trial. Wage effects were quanti-fied as the average change in work time between a baseline trialand the associated wage-change trials. We predicted that bothwage manipulations would again cause participants to forgo theirbaseline time-allocation choices, leading them to sacrifice incomein the case of wage decreases and to sacrifice leisure in the case ofwage increases.

Analyses are described in conjunction with results. Tests onwage effects excluded one participant falling more than two stan-dard deviations from the sample mean, on either wage increase ordecrease trials.

Results and Discussion

Table 1 lists mean response accuracies and d= scores for the low-and high-demand tasks, which as in previous studies were wellabove chance (ps � 0.001), indicating task compliance.

Mean leisure time and income before and after wage changesare shown in Figure 2C (for wage decreases) and 2D (increases).

Thi

sdo

cum

ent

isco

pyri

ghte

dby

the

Am

eric

anPs

ycho

logi

cal

Ass

ocia

tion

oron

eof

itsal

lied

publ

ishe

rs.

Thi

sar

ticle

isin

tend

edso

lely

for

the

pers

onal

use

ofth

ein

divi

dual

user

and

isno

tto

bedi

ssem

inat

edbr

oadl

y.

136 KOOL AND BOTVINICK

Across participants and trials, income-compensated wage de-creases triggered a 29% increase in leisure and a 12% decrease inincome, t(17) � 3.41, p � .05; Cohen’s d � 0.80, while income-compensated wage increases triggered a 15% decrease in leisure

and an 11% increase in income, t(17) � 2.79, p � .05, Cohen’sd � 0.66.

Interestingly, the “elasticity” of time allocation—that is, itssensitivity to a unit change in wage—appeared to differ reliablyacross participants. The degree to which each participant reducedwork in the face of wage reductions correlated with the degree towhich they increased work in response to a boost in wages (r �.68, p � .001; see Figure 4). Further analyses suggested that flooror ceiling effects were not wholly responsible for this correlation;the size of wage-change effects did not significantly correlate withmean leisure-time allocation (r � .31, p � .20) or with the absolutedeviation of leisure-time allocation from 50% (r � –.39, p � .11).

Beyond its primary goals, the present study also provided anopportunity to address a lingering concern from Experiments 1 and2. In the Experiments 1 and 2: Additional Analyses section, wecompared cases where wage was matched but effort allocationdiffered, taking the difference as evidence that participants tookunearned income into account. However, the sessions compared inthat analysis differed not only in terms of unearned income but interms of context: One of the sessions (Session 2 from Experiment2) was preceded by a session involving a lower wage, while theother (Session 1 from Experiment 1) was not. The difference ineffort allocation could thus, arguably, have arisen from an effect ofcontext on the evaluation of wage. Specifically, the wage offeredin Experiment 2, Session 2, might have appeared more rich andmotivating, given the contrast with the earlier, lower wage. Inorder to evaluate this possibility, as well as to further validate therole of unearned income, we conducted a linear regression for eachparticipant in Experiment 3, modeling the chosen leisure time oneach trial as a function of (1) the present wage, (2) the present

M&M’s M&M’s

M&M’s M&M’s M&M’s M&M’s0.0

0.0

5.3

12.3

20.0

20.0M&M’s

12.3

M&M’s’ M&M’s’ M&M’s’

12.3

M&M’s’

M&M’s’M&M’s’

12.3

A B C

D E F G

Choice

Choice

Figure 3. In the choice phase of Experiment 3, the graphical interface showed a single pie chart indicating labor(red) and leisure (green) time, along with an associated total income (in pieces of candy). Participants explored therange of available income/leisure combinations, using arrow keys to increase or decrease leisure time in smallincrements (curved arrow, not shown in user interface). Panels A and C in the figure show the extreme allocationoptions on one choice trial, and B shows the combination actually chosen by the participant. In the lower row, PanelsD and G show the extreme allocation options on another trial faced by the same participant, which implemented anincome-compensated wage decrease relative to the trial diagrammed in the top row. Although the income/leisurecombination from Panel B was available to the participant on this trial (Panel F), the participant chose a combinationinvolving greater leisure and less income (Panel E), consistent with the predictions of labor supply theory.

Table 2Trial-Type Parameters From Experiment 3

Group Wage Unearned income

A 3.6 None3.6 � 1/2 Compensatory (�0)3.6 � 1/3 Compensatory (�0)3.6 � 1/4 Compensatory (�0)

B 1.8 None1.8 � 2/3 Compensatory (�0)1.8 � 1/2 Compensatory (�0)1.8 � 2/5 Compensatory (�0)

C 1.2 481.2 � 1.5 Compensatory (�48)1.2 � 2 Compensatory (�48)1.2 � 3 Compensatory (�48)

D 0.6 360.6 � 2 Compensatory (�36)0.6 � 3 Compensatory (�36)0.6 � 4 Compensatory (�36)

Note. Trials were divided into four groups, each containing four specifictrial types. Each group included one parameterization that served as abaseline condition for the others (first row in each group). The remainingthree trial types each implemented an income-compensated wage decrease(Groups A and B) or increase (Groups C and D) with respect to the relevantbaseline. Each trial type formed the basis for four actual trials for eachparticipant. Wages are designated in units of candy pieces per cumulativeminute of labor.

Thi

sdo

cum

ent

isco

pyri

ghte

dby

the

Am

eric

anPs

ycho

logi

cal

Ass

ocia

tion

oron

eof

itsal

lied

publ

ishe

rs.

Thi

sar

ticle

isin

tend

edso

lely

for

the

pers

onal

use

ofth

ein

divi

dual

user

and

isno

tto

bedi

ssem

inat

edbr

oadl

y.

137COGNITIVE LABOR/LEISURE DECISIONS

unearned income, and (3) the difference between present wage andthe wage offered on the immediately preceding trial (see Table 3).We tested whether each of these regressors, on average, explainedvariance in participants’ choice behavior. Leisure time was, notsurprisingly, predicted by present wage (p � .01). More important,it was also predicted by unearned income (p � .05) but not reliablypredicted by wage difference (p � .41). This pattern of resultsfurther supports the integral role of income compensation in driv-ing the main effects from all three studies.

In order to rule out a key role for error avoidance, we leveragedthe data from the initial practice session. We reasoned that if effortallocation were based entirely on error rates rather than cognitivedemand, then leisure time on baseline trials should correlate acrossparticipants with accuracy on the high-demand task during thepractice block. However, no such correlation was observed, re-gardless of whether accuracy was quantified as d=, proportioncorrect, or proportion correct only on target trials (p � .80 in allcases).

General Discussion

To summarize, we found across three experiments that income-compensated wage changes impacted cognitive labor/leisure deci-

sions in a fashion consistent with economic labor supply theory:Even when participants had the option to recreate a previouslyselected combination of income and cognitive leisure, wage re-ductions led them to work less and settle for a smaller income, andwage increases led them to give up leisure in favor of a largerincome. The observed pattern of behavior suggests that cognitivelabor/leisure decisions are guided by preferences that jointly weighincome and leisure.

Our findings directly parallel results from both experimentallabor economics (Charness & Kuhn, 2010; Fehr, Goette, &Zehnder, 2009) and animal conditioning research (Chen et al.,2006; Conover & Shizgal, 2005; Kagel et al., 1995), indicating thatthe principles that guide effort allocation both in labor markets andin animal foraging may also apply in the more rarified domain ofhuman cognitive control.

On the broader stage, effects of incentives on cognitive effortand performance have been of interest both in basic-science set-tings (Brehm, Wright, Solomon, Silka, & Greenberg, 1983; Koolet al., 2010; Ryan & Deci, 2000b; Sarter, Gehring, & Kozak, 2006)and in applied settings including education (Ryan & Deci, 2000a;Weiner, 1985). Our work suggests that, within such work, it maybe fruitful to apply formal tools from labor economics. For exam-ple, economic research has demonstrated nonmonotonic effects ofincentives on worker effort (resulting in what is known as thebackward-bending labor supply curve) and interactions betweenwage incentives and social preferences on worker behavior (Cam-erer, Babcock, Loewenstein, & Thaler, 1997). The present findingsopen up the question of whether effects such as these might alsocharacterize decision making in cognitive control.

In addition to work on incentive effects, the present findingsalso relate closely to research on the cost of cognitive control. Anumber of recent studies (Botvinick, 2007; Botvinick, Huffstetler,& McGuire, 2009; Kool et al., 2010; McGuire & Botvinick, 2010)have lent empirical support to the long-standing proposition thatthe exertion of executive control is intrinsically costly or aversive.This principle is evident, for example, in choice tasks whereparticipants are free to select between two lines of action associ-ated with different levels of cognitive demand. Human decisionmakers in this setting favor the low-demand option, a demand-avoidance effect that (like some of the present findings) is notcompletely attributable to ancillary factors such as error avoidanceor time on task (Kool et al., 2010).

The demand-avoidance effect fits into the framework we havebeen developing here via a simple change of sign: In cognitivelabor/leisure decisions, the utility of leisure derives, in importantpart, from the relief it offers from costly control. Equivalently, onemay view the cost of control as an opportunity cost, relating to

Table 3Regression Coefficients and Statistics for the Regression Analysis Performed in Experiment 3

MeasureRegressor currentunearned income Current wage �Wagecurrent–previous

Regression coefficient (SE) �0.04 (0.02) 1.34 (0.42) 0.12 (0.15)t (df) –2.22 (17) 3.20 (17) 0.85 (17)p �.05 �.01 .41

Note. For each participant, we modeled the time allocated to labor on each trial as a function of the current wage, current unearned income, and thedifference in wage between the current and the previous trials.

Resp

on

se t

o w

age

incr

ease

Response to wage decrease

r = 0.68**

-6

11

-6 11

Figure 4. Individual-participant data from Experiment 3. The depictedcorrelation indicates that, when participants showed a large shift to leisureafter a compensated wage decrease, they also showed a large shift to laborafter a compensated wage increase. There was significant variability in thismeasure of elasticity, or the sensitivity to a unit change in wage, acrossparticipants.

Thi

sdo

cum

ent

isco

pyri

ghte

dby

the

Am

eric

anPs

ycho

logi

cal

Ass

ocia

tion

oron

eof

itsal

lied

publ

ishe

rs.

Thi

sar

ticle

isin

tend

edso

lely

for

the

pers

onal

use

ofth

ein

divi

dual

user

and

isno

tto

bedi

ssem

inat

edbr

oadl

y.

138 KOOL AND BOTVINICK

forgone leisure. The beauty of the labor supply model is that itrenders the choice between these two perspectives moot. Theshape of the utility function says it all.

Additional studies of the demand-avoidance effect have shownthat it is reduced in the presence of countervailing incentives,indicating that the inherent cost of control is weighed againstpotential rewards (Kool et al., 2010). The present work provides aricher picture of this cost–benefit analysis. By default, one mightassume a simple linear model, under which the net value of acourse of action is computed by subtracting the cost of controlfrom associated rewards. The present findings suggest, instead,that the marginal cost of control varies as a function of context: Aunit increment in effort carries a greater subjective cost when oneis already working hard than when one is hardly working.

This way of characterizing our results reveals a natural relation-ship with research on the phenomenon of ego depletion. The termrefers to the observation that voluntary cognitive effort tends todecline after bouts of forced cognitive exertion (Baumeister, 2002;Baumeister, Bratslavsky, Muraven, & Tice, 1998). According toone influential theory, this effect reflects a depletion of bloodglucose, a metabolic resource that is hypothesized to be necessaryfor the exertion of cognitive control (Baumeister, Vohs, & Tice,2007; Gailliot et al., 2007). The present work suggests an alterna-tive view of the depletion effect, according to which it arises froma particular form of cost–benefit analysis: In the context of pro-longed, obligatory mental effort, the marginal cost of further effortis elevated, leading in some cases to a subsequent withdrawal fromcognitively challenging activity. By characterizing the depletioneffect in this way, the present work situates this specific phenom-enon within a larger framework for understanding labor/leisuredecisions.

Unlike the glucose depletion account, the present theory isprimarily descriptive rather than mechanistic or normative. Ourgoal has been to account for behavior in terms of an underlyingutility function, the same strategy that is used by theories ofeconomic choice (e.g., prospect theory) in approaching phenomenasuch as loss aversion or risk seeking (Kahneman & Tversky,1979). This descriptive orientation may appear to place the presentwork at a disadvantage, when compared with the more mechanisticresource-depletion model. In response to this potential challenge,two points may be offered. First, it should be noted that empiricalevidence for the role of glucose (or any other metabolic resource)in the behavioral “depletion” effect is, at best, mixed (Kurzban,2010), undermining depletion’s force as a mechanistic argument.Second, a range of findings—for example, the disappearance ofdepletion effects in the face of increased incentives (Muraven &Slessareva, 2003)—has led even proponents of the ego-depletiontheory to acknowledge the need for an additional layer, at whicheffort expenditure is shaped by motivational factors (Hagger,Wood, Stiff, & Chatzisarantis, 2010). The present account, whichis pitched precisely at this level, thus fills a gap left even bysupposedly mechanistic approaches.

Another appealing aspect of the ego-depletion theory is itsnormative flavor. The theory appears to make clear why it may berational to withdraw cognitive control in certain contexts: Suchwithdrawal allows for the preservation or replenishment of alimited metabolic resource (viz., glucose; Muraven, Shmueli, &Burkley, 2006). The present account, as a descriptive theory, doesnot center on (or depend upon) a normative argument. Having said

this, it is possible, if one is willing to speculate, to fit the presentdescriptive theory within a normative framework. From a compu-tational point of view, there are in fact two reasons why theoccasional withdrawal of control may be rational and why it mightthus be rational for decision making to place a cost on control, aswe have proposed. First, a wealth of research has indicated thatcontrolled information processing is highly capacity-limited. Thus,a motivational bias in favor of automatic processing would havethe effect of reserving this limited computational bandwidth foroperations that are likely to have a high payoff (see Kool et al.,2010). Second, and more speculatively, it seems plausible that theleisure state arising when control is disengaged may allow thedecision maker to survey or explore available activities, preventinga myopic focus on one potentially suboptimal line of behavior(Kool et al., 2010). In this sense, the labor/leisure tradeoff dem-onstrated in our experiments may address a high-level instantiationof the exploitation/exploration tradeoff that arises inreinforcement-learning models of reward-based decision making(Cohen, McClure, & Yu, 2007) and of foraging (Kolling, Behrens,Mars, & Rushworth, 2012).

We began this article with an imagined real-life example oflabor/leisure decision making. The question may now be asked,How do the results of our experiments relate to everyday life? Twoaspects of our experimental approach call for comment, in thisconnection. First, the decisions made by participants in our exper-iments were deliberate and explicit, a feature that allowed us tointerpret them as reflecting participants’ actual preferences. Sec-ond, the labor/leisure decision in our experiments involved select-ing between two active tasks, a measure that allowed us to matchlabor and leisure conditions in terms of stimulus and responsecharacteristics, physical effort demands, and intrinsic task interest.Some labor/leisure decisions in everyday life clearly involve adeliberate choice between two active tasks. Our earlier homework-versus-video example fits this description, as do certain casesinvolving failures of self-control, where one throws off the burdenof self-discipline (read: cost of control) in order to indulge in aformerly resisted temptation (Kool, McGuire, Wang, & Botvinick,2012). At the same time, intuition suggests that many everydaylabor/leisure decisions are unreflective, or even unconscious, andpertain to the level of engagement in a single task rather than to achoice between tasks. The phenomenon of mind wandering, inparticular, seems relevant here, since it is conventionally under-stood as involving a nondeliberate disengagement from a single,focal task. The theoretical framework we have proposed doesplausibly extend to this scenario. On a labor-supply account, mindwandering would be understood as an implicit, unreflective choiceto engage in cognitive leisure (here, avoidance of the focal task infavor of nondemanding but otherwise unspecified mental activi-ties), triggered by an elevation of marginal control costs above taskpayoffs. Mind wandering has been shown to reflect a disengage-ment of cognitive control (McVay & Kane, 2012) and, interest-ingly, to be accompanied by increased activity in the default-modebrain network (Christoff, Gordon, Smallwood, Smith, & Schooler,2009). However, further experimentation will be required to es-tablish whether the labor-supply framework we have introducedindeed applies to this and other variants of cognitive labor/leisuredecision making.

What neural mechanisms might underlie the pattern of labor/leisure decision making observed in our experiments? The exertion

Thi

sdo

cum

ent

isco

pyri

ghte

dby

the

Am

eric

anPs

ycho

logi

cal

Ass

ocia

tion

oron

eof

itsal

lied

publ

ishe

rs.

Thi

sar

ticle

isin

tend

edso

lely

for

the

pers

onal

use

ofth

ein

divi

dual

user

and

isno

tto

bedi

ssem

inat

edbr

oadl

y.

139COGNITIVE LABOR/LEISURE DECISIONS

of cognitive control is well known to engage a specific set of areaswithin dorsolateral prefrontal cortex, dorsomedial frontal cortex,the insula, and parietal cortex (Duncan, 2010). Recent evidencehas indicated that portions of this network also index, in tandemwith related subcortical structures, both the potential payoffs (Kou-neiher, Charron, & Koechlin, 2009; Locke & Braver, 2008) andthe subjective cost of control (Botvinick et al., 2009; McGuire &Botvinick, 2010). As noted earlier, a parallel literature has pointedto a neural correlate of cognitive leisure, linking the disengage-ment of cognitive control with activation of a default mode net-work, centering on ventral prefrontal cortex (Kelly, Uddin, Biswal,Castellanos, & Milham, 2008; Raichle et al., 2001). We speculatethat the labor/leisure tradeoff demonstrated in the present behav-ioral studies reflects the operation of a neural mechanism regulat-ing the balance of activity between these dorsal “task-positive” andventromedial “task-negative” systems (Menon & Uddin, 2010;Sridharan, Levitin, & Menon, 2008). If this is correct, then theformal framework we have introduced here might usefully informfuture neuroscientific research into such a mediating neural mech-anism.

References

Baler, R. D., & Volkow, N. D. (2006). Drug addiction: The neurobiologyof disrupted self-control. Trends in Molecular Medicine, 12, 559–566.doi:10.1016/j.molmed.2006.10.005

Baumeister, R. F. (2002). Ego depletion and self-control failure: An energymodel of the self’s executive function. Self and Identity, 1, 129–136.doi:10.1080/152988602317319302

Baumeister, R. F., Bratslavsky, E., Muraven, M., & Tice, D. M. (1998).Ego depletion: Is the active self a limited resource? Journal of Person-ality and Social Psychology, 74, 1252–1265. doi:10.1037/0022-3514.74.5.1252

Baumeister, R. F., Vohs, K. D., & Tice, D. M. (2007). The strength modelof self-control. Current Directions in Psychological Science, 16, 351–355. doi:10.1111/j.1467-8721.2007.00534.x

Bigelow, G., & Liebson, I. (1972). Cost factors controlling alcoholicdrinking. Psychological Record, 22, 305–314.

Bonner, S. E., & Sprinkle, G. B. (2002). The effects of monetary incentiveson effort and task performance: Theories, evidence, and a framework forresearch. Accounting, Organizations and Society, 27, 303–345. doi:10.1016/S0361-3682(01)00052-6

Botvinick, M. M. (2007). Conflict monitoring and decision making: Rec-onciling two perspectives on anterior cingulate function. Cognitive,Affective, & Behavioral Neuroscience, 7, 356–366. doi:10.3758/CABN.7.4.356

Botvinick, M. M., Huffstetler, S., & McGuire, J. T. (2009). Effort dis-counting in human nucleus accumbens. Cognitive, Affective, & Behav-ioral Neuroscience, 9, 16–27. doi:10.3758/CABN.9.1.16

Braver, T. S., Cohen, J. D., Nystrom, L. E., Jonides, J., Smith, E. E., &Noll, D. C. (1997). A parametric study of prefrontal cortex involvementin human working memory. NeuroImage, 5, 49–62. doi:10.1006/nimg.1996.0247

Brehm, J. W., Wright, R. A., Solomon, S., Silka, L., & Greenberg, J.(1983). Perceived difficulty, energization, and the magnitude of goalvalence. Journal of Experimental Social Psychology, 19, 21–48. doi:10.1016/0022-1031(83)90003-3

Camerer, C. F., Babcock, L., Loewenstein, G., & Thaler, R. (1997). Laborsupply of New York City cabdrivers: One day at a time. QuarterlyJournal of Economics, 112, 407–441. doi:10.1162/003355397555244

Camerer, C. F., & Hogarth, R. M. (1999). The effects of financial incen-tives in experiments: A review and capital-labor-production framework.

Journal of Risk and Uncertainty, 19, 7– 42. doi:10.1023/A:1007850605129

Charness, G., & Kuhn, P. (2010). Lab labor: What can labor economistslearn from the lab? (IZA Discussion Paper No. 4941). Bonn, Germany:Institute for the Study of Labor.

Chen, M. K., Lakshminarayanan, V., & Santos, L. R. (2006). How basicare behavioral biases? Evidence from capuchin monkey trading behav-ior. Journal of Political Economy, 114, 517–537. doi:10.1086/503550.

Christoff, K., Gordon, A. M., Smallwood, J., Smith, R., & Schooler, J. W.(2009). Experience sampling during fMRI reveals default network andexecutive system contributions to mind wandering. PNAS Proceedingsof the National Academy of Sciences, USA, 106, 8719–8724. doi:10.1073/pnas.0900234106

Cohen, J. D., McClure, S. M., & Yu, A. J. (2007). Should I stay or shouldI go? How the human brain manages the trade-off between exploitationand exploration. Philosophical Transactions of the Royal Society B:Biological Sciences, 362, 933–942. doi:10.1098/rstb.2007.2098

Collier, G., & Jennings, W. (1969). Work as a determinant of instrumentalperformance. Journal of Comparative and Physiological Psychology,68, 659–662. doi:10.1037/h0027670

Conover, K. L., & Shizgal, P. (2005). Employing labor-supply theory tomeasure the reward value of electrical brain stimulation. Games andEconomic Behavior, 52, 283–304. doi:10.1016/j.geb.2004.08.003

Dickinson, D. L. (1999). An experimental examination of labor supply andwork intensities. Journal of Labor Economics, 17, 638–670. doi:10.1086/209934

Duncan, J. (2010). The multiple-demand (MD) system of the primate brain:Mental programs for intelligent behaviour. Trends in Cognitive Sciences,14, 172–179. doi:10.1016/j.tics.2010.01.004

Eichele, T., Debener, S., Calhoun, V., Specht, K., Engel, A., Hugdahl, K.,. . . Ullsperger, M. (2008). Prediction of human errors by maladaptivechanges in event-related brain networks. PNAS Proceedings of theNational Academy of Sciences, USA, 105, 6173–6178. doi:10.1073/pnas.0708965105

Fehr, E., & Goette, L. (2007). Do workers work more if wages are high?Evidence from a randomized field experiment. American EconomicReview, 97, 298–317. doi:10.1257/aer.97.1.298

Fehr, E., Goette, L., & Zehnder, C. (2009). A behavioral account of thelabor market: The role of fairness concerns. Annual Review of Econom-ics, 1, 355–384. doi:10.1146/annurev.economics.050708.143217

Gailliot, M. T., Baumeister, R. F., DeWall, C. N., Maner, J. K., Plant,E. A., Tice, D. M.,. . . Schmeichel, B. J. (2007). Self-control relies onglucose as a limited energy source: Willpower is more than a metaphor.Journal of Personality and Social Psychology, 92, 325–336. doi:10.1037/0022-3514.92.2.325.

Green, L., Kagel, J. H., & Battalio, R. C. (1987). Consumption-leisuretradeoffs in pigeons: Effects of changing marginal wage rates by varyingamount of reinforcement. Journal of the Experimental Analysis of Be-havior, 47, 17–28. doi:10.1901/jeab.1987.47-17

Hagger, M. S., Wood, C., Stiff, C., & Chatzisarantis, N. L. D. (2010). Egodepletion and the strength model of self-control: A meta-analysis. Psy-chological Bulletin, 136, 495–525. doi:10.1037/a0019486

Jimura, K., Locke, H. S., & Braver, T. S. (2010). Prefrontal cortexmediation of cognitive enhancement in rewarding motivational contexts.PNAS Proceedings of the National Academy of Sciences, USA, 107,8871–8876. doi:10.1073/pnas.1002007107

Kagel, J. H., Battalio, R. C., & Green, L. (1995). Economic choice theory:An experimental analysis of animal behavior. Cambridge, England:Cambridge University Press. doi:10.1017/CBO9780511664854

Kahneman, D., & Tversky, A. (1979). Prospect theory: An analysis ofdecision under risk. Econometrica, 47, 263–292. doi:10.2307/1914185

Kelly, A. M., Uddin, L. Q., Biswal, B. B., Castellanos, F. X., & Milham,M. P. (2008). Competition between functional brain networks mediates

Thi

sdo

cum

ent

isco

pyri

ghte

dby

the

Am

eric

anPs

ycho

logi

cal

Ass

ocia

tion

oron

eof

itsal

lied

publ

ishe

rs.

Thi

sar

ticle

isin

tend

edso

lely

for

the

pers

onal

use

ofth

ein

divi

dual

user

and

isno

tto

bedi

ssem

inat

edbr

oadl

y.

140 KOOL AND BOTVINICK

behavioral variability. NeuroImage, 39, 527–537. doi:10.1016/j.neuroimage.2007.08.008

Kolling, N., Behrens, T. E. J., Mars, R. B., & Rushworth, M. F. S. (2012,April 6). Neural mechanisms of foraging. Science, 336, 95–98. doi:10.1126/science.1216930

Kool, W., McGuire, J. T., Rosen, Z. B., & Botvinick, M. M. (2010).Decision making and the avoidance of cognitive demand. Journal ofExperimental Psychology: General, 139, 665– 682. doi:10.1037/a0020198

Kool, W., McGuire, J. T., Wang, G. J., & Botvinick, M. M. (2012). A rolefor effort costs in self-control and intertemporal choice. Manuscriptsubmitted for publication.

Kouneiher, F., Charron, S., & Koechlin, E. (2009). Motivation and cogni-tive control in the human prefrontal cortex. Nature Neuroscience, 12,939–945. doi:10.1038/nn.2321

Kurzban, R. (2010). Does the brain consume additional glucose duringself-control tasks? Evolutionary Psychology, 8, 244–259.

Locke, H. S., & Braver, T. S. (2008). Motivational influences on cognitivecontrol: Behavior, brain activation, and individual differences. Cogni-tive, Affective, & Behavioral Neuroscience, 8, 99–112. doi:10.3758/CABN.8.1.99

McGuire, J. T., & Botvinick, M. M. (2010). Prefrontal cortex, cognitivecontrol, and the registration of decision costs. PNAS Proceedings of theNational Academy of Sciences USA, 107, 7922–7926. doi:10.1073/pnas.0910662107

McVay, J. C., & Kane, M. J. (2012). Drifting from slow to “d’oh!”:Working memory capacity and mind wandering predict extreme reactiontimes and executive control errors. Journal of Experimental Psychology:Learning, Memory, and Cognition, 38, 525–549. doi:10.1037/a0025896

Menon, V., & Uddin, L. Q. (2010). Saliency, switching, attention andcontrol: A network model of insula function. Brain Structure & Func-tion, 214, 655–667. doi:10.1007/s00429-010-0262-0.

Miller, E. K., & Cohen, J. D. (2001). An integrative theory of prefrontalcortex function. Annual Review of Neuroscience, 24, 167–202.

Muraven, M., Shmueli, D., & Burkley, E. (2006). Conserving self-controlstrength. Journal of Personality and Social Psychology, 91, 524–537.doi:10.1037/0022-3514.91.3.524

Muraven, M., & Slessareva, E. (2003). Mechanisms of self-control failure:Motivation and limited resources. Personality and Social PsychologyBulletin, 29, 894–906. doi:10.1177/0146167203029007008

Muraven, M., Tice, D. M., & Baumeister, R. F. (1998). Self-control aslimited resource: Regulatory depletion patterns. Journal of Personalityand Social Psychology, 74, 774–789. doi:10.1037/0022-3514.74.3.774

Murphy, F. C., Rubinsztein, J. S., Michael, A., Rogers, R. D., Robbins,T. W., Paykel, E. S., & Sahakian, B. J. (2001). Decision-making cog-nition in mania and depression. Psychological Medicine, 31, 679–693.doi:10.1017/S0033291701003804

Nicholson, W., & Snyder, C. M. (2008). Microeconomic theory: Basicprinciples and extensions. Mason, OH: Cengage Learning.

Raichle, M. E., MacLeod, A. M., Snyder, A. Z., Powers, W. J., Gusnard,D. A., & Shulman, G. L. (2001). A default mode of brain function. PNASProceedings of the National Academy of Sciences, USA, 98, 676–682.doi:10.1073/pnas.98.2.676

Ryan, R. M., & Deci, E. L. (2000a). Intrinsic and extrinsic motivations:Classic definitions and new directions. Contemporary Educational Psy-chology, 25, 54–67. doi:10.1006/ceps.1999.1020

Ryan, R. M., & Deci, E. L. (2000b). Self-determination theory and thefacilitation of intrinsic motivation, social development, and well-being.American Psychologist, 55, 68–78. doi:10.1037/0003-066X.55.1.68

Sarter, M., Gehring, W. J., & Kozak, R. (2006). More attention must bepaid: The neurobiology of attentional effort. Brain Research Reviews,51, 145–160. doi:10.1016/j.brainresrev.2005.11.002

Sridharan, D., Levitin, D., & Menon, V. (2008). A critical role for the rightfronto-insular cortex in switching between central-executive and default-mode networks. PNAS Proceedings of the National Academy of Sci-ences, USA, 105, 12569–12574. doi:10.1073/pnas.0800005105

van der Plas, E. A. A., Crone, E. A., van den Wildenberg, W. P. M., Tranel,D., & Bechara, A. (2009). Executive control deficits in substance-dependent individuals: A comparison of alcohol, cocaine, and metham-phetamine and of men and women. Journal of Clinical and ExperimentalNeuropsychology, 31, 706–719. doi:10.1080/13803390802484797

Weiner, B. (1985). An attributional theory of achievement motivation andemotion. Psychological Review, 92, 548–573. doi:10.1037/0033-295X.92.4.548

Weissman, D. H., Roberts, K. C., Visscher, K. M., & Woldorff, M. G.(2006). The neural bases of momentary lapses in attention. NatureNeuroscience, 9, 971–978. doi:10.1038/nn1727

Wenzlaff, R. M., & Bates, D. E. (1998). Unmasking a cognitive vulnera-bility to depression: How lapses in mental control reveal depressivethinking. Journal of Personality and Social Psychology, 75, 1559–1571.doi:10.1037/0022-3514.75.6.1559

Received December 1, 2011Revision received October 24, 2012

Accepted October 26, 2012 �

Thi

sdo

cum

ent

isco

pyri

ghte

dby

the

Am

eric

anPs

ycho

logi

cal

Ass

ocia

tion

oron

eof

itsal

lied

publ

ishe

rs.

Thi

sar

ticle

isin

tend

edso

lely

for

the

pers

onal

use

ofth

ein

divi

dual

user

and

isno

tto

bedi

ssem

inat

edbr

oadl

y.

141COGNITIVE LABOR/LEISURE DECISIONS